def main():
# 0. read data and splite it to 80% for training and 20% for testing
items = pd.read_csv('input/items.csv', sep=';', encoding='ISO-8859-1')
print items.shape
items_train, items_test = train_test_split(items, train_size=0.8, random_state=0)
print items_train.shape, items_test.shape
# 1. train tf-idf model and save it under model/tf-idf-model.pickle with the result
if not os.path.isfile('model/tfidf_model.pickle'):
print('traning tf-idf model ...')
tfidf_model = TfidfVectorizer(norm='l2',min_df=0, use_idf=True,max_features=5000, smooth_idf=False, sublinear_tf=True, tokenizer=tokeniser)
item_feature_matrix = tfidf_model.fit_transform(items_train['movie desription'].values.astype('U'))
print('#1. dimension of the item-feature matrix', item_feature_matrix.shape)
# 1.1 saving tf-idf model
print('Saving tf-idf model ...')
save_model('model/tfidf_model.pickle', tfidf_model)
if not os.path.isfile('result/item_feature_matrix.pickle'):
# 1.2. saving tf-idf matrix result
print('Saving tf-idf matrix result ...')
save_model('result/item_feature_matrix.pickle', item_feature_matrix)
# 2. train dbn model and save the model into model/dbn.pickle
# 2.1. load tf-idf result
print('loading item feature matrix ...')
item_feature_matrix = load_model('result/item_feature_matrix.pickle')
if not os.path.isfile('model/dbn-model.pkl'):
dbn = UnsupervisedDBN(hidden_layers_structure=[5000, 400],
batch_size=10,
learning_rate_rbm=0.06,
n_epochs_rbm=20,
activation_function='sigmoid')
# 2.2. fit dbn model
dbn.fit(item_feature_matrix.A)
# 2.3. save dbn model
print('saving DBN model ...')
dbn.save('model/dbn-model.pkl')
print('Loadin DBN model')
dbn = UnsupervisedDBN.load('model/dbn-model.pkl')
# 3. Clustering with k-mens and save model and results
if not os.path.isfile('model/kmeans-model.pkl'):
kmeans = KMeans(n_clusters=5, random_state=0).fit(dbn.transform(item_feature_matrix.A))
print('saving k-means model ...')
save_model('model/kmeans-model.pkl', kmeans)
else:
kmeans = load_model('model/kmeans-model.pkl')
print(kmeans.labels_)
header=None,
encoding='ISO-8859-1')
u_item_DF['movie desription'] = [val[2] for i, val in data_new.iterrows()]
sklearn_tfidf = TfidfVectorizer(norm='l2',
min_df=0,
use_idf=True,
max_features=5000,
smooth_idf=False,
sublinear_tf=True,
tokenizer=tokeniser)
item_feature_matrix = sklearn_tfidf.fit_transform(
u_item_DF['movie desription'].values.astype('U'))
print('dimension of the item-feature matrix', item_feature_matrix.shape)
# Train DBN model
from dbn.models import UnsupervisedDBN
#[4604, 2000, 4000, 3000, 1000]
dbn = UnsupervisedDBN(hidden_layers_structure=[5000, 400],
batch_size=10,
learning_rate_rbm=0.06,
n_epochs_rbm=20,
activation_function='sigmoid')
dbn.fit(item_feature_matrix.A)
# Save the model
print('Saving Model ...')
dbn.save('model-1.pkl')
print('Model Saved')
def main(tfidfModel=None, tfidfMatrix=None, dbn_model=None, kmeans_model=None):
# 0. read data and splite it to 80% for training and 20% for testing
items_info = pd.read_csv('input/items.csv', sep=';', encoding='ISO-8859-1')
u_base1 = pd.read_csv('input/u1.base', sep='\t', header=None)
train = pd.DataFrame(u_base1[1].drop_duplicates())
u_test1 = pd.read_csv('input/u1.test', sep='\t', header=None)
test = pd.DataFrame(u_test1[1].drop_duplicates())
train_desc = [
items_info[items_info['movie id'] == df[1]]
['movie desription'].values[0] for i, df in train.iterrows()
]
test_desc = [
items_info[items_info['movie id'] == df[1]]
['movie desription'].values[0] for i, df in test.iterrows()
]
# 1. train tf-idf model and save it under model/tf-idf-model.pickle with the result
if not tfidfModel:
print('traning tf-idf model ...')
tfidf_model = TfidfVectorizer(norm='l2',
min_df=0,
use_idf=True,
max_features=5000,
smooth_idf=False,
sublinear_tf=True,
tokenizer=tokeniser)
tfidf_model.fit(train_desc)
print('- Saving tf-idf model ...')
save_model('model/tfidf_model.pickle', tfidf_model)
else:
print('# Loading tf-idf model ...')
tfidf_model = load_model(tfidfModel)
if not tfidfMatrix:
item_feature_matrix = tfidf_model.transform(train_desc)
# 1.2. saving tf-idf matrix result
print('- Saving tf-idf matrix result ...')
save_model('result/item_feature_matrix.pickle', item_feature_matrix)
else:
print('# Loading tf-idf matrix result ...')
item_feature_matrix = load_model(tfidfMatrix)
if not dbn_model:
dbn = UnsupervisedDBN(hidden_layers_structure=[5000, 1000, 1000, 500],
batch_size=10,
learning_rate_rbm=0.06,
n_epochs_rbm=20,
activation_function='sigmoid')
# 2.2. fit dbn model
dbn.fit(item_feature_matrix.A)
# 2.3. save dbn model
print('saving DBN model ...')
dbn.save('model/dbn-model.pkl')
else:
print('Loadin DBN model')
dbn = UnsupervisedDBN.load(dbn_model)
# 3. Clustering with k-mens and save model and results
if not kmeans_model:
kmeans = KMeans(n_clusters=5, random_state=0).fit(
dbn.transform(item_feature_matrix.A))
print('saving k-means model ...')
save_model('model/kmeans-model.pkl', kmeans)
else:
print('loading k-means model ...')
kmeans = load_model(kmeans_model)
print("Done!")